Computer vision for automatic identification of blastocyst structures and blastocyst formation time in In-Vitro...

Area of Science:

• Biomedical Engineering
• Reproductive Medicine
• Computer Vision

Background:

• Embryo selection for In-Vitro Fertilization (IVF) is crucial but subjective and time-consuming.
• Current IVF success rates remain suboptimal, with a 34.1% pregnancy rate even with top-quality embryos.
• Objective embryo analysis is needed to reduce embryologist workload and improve implantation outcomes.

Purpose of the Study:

• To develop and present computer vision methods for automatic segmentation of human blastocyst structures.
• To accurately determine the timing of expanded blastocyst formation.
• To enhance the reliability and comparability of blastocyst analysis in assisted reproduction.

Main Methods:

• Implementation of various computer vision algorithms for image analysis.
• Automatic segmentation of key structures within the blastocyst.
• Validation of segmentation accuracy using the Dice Score metric.
• Precise temporal identification of expanded blastocyst stage.

Main Results:

• Achieved high accuracy in blastocyst structure segmentation, with a Dice Score up to 0.89.
• Accurately identified the expanded blastocyst formation timing with a mean error under 4 hours.
• Developed objective, quantitative data for embryo assessment.

Conclusions:

• Computer vision offers a valuable tool for objective and efficient blastocyst analysis in IVF.
• Automated segmentation and timing can significantly aid embryologists and improve IVF success rates.
• Open-sourced code facilitates replication and further research in assisted reproduction technology.